Kyutai TTS-integratie

Voer Kyutai-modellen lokaal uit. Deze gids behandelt Pocket TTS (CPU) en TTS 1.6B (GPU) met meegeleverde OpenAI-compatibele servers.

Overzicht

Model	Parameters	Apparaat	Geschikt voor
Pocket TTS	100M	Alleen CPU	Laptops, lichte omgevingen
TTS 1.6B	1.6B	GPU/MPS/CPU	Servers, hoge kwaliteit

Beide gebruiken CALM (Continuous Audio Language Models) en klonen vanaf audio.

Pocket TTS (CPU)

Realtime op CPU, zonder GPU.

Vereisten

Onderdeel	Minimum
Python	3.10 - 3.14
PyTorch	2.5+
RAM	4GB
Schijf	500MB

Snel starten

cd examples/kyutai-tts-server

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Start server
python server.py

Server: http://localhost:8200.

Test

curl http://localhost:8200/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model": "kyutai-tts", "input": "Hello, welcome to Libre WebUI!", "voice": "alba"}' \
  --output speech.wav

Stemmen

Stem	Beschrijving
Alba	Vrouwelijk, helder/natuurlijk
Marius	Mannelijk, warm
Javert	Mannelijk, gezaghebbend
Jean	Mannelijk, zacht
Fantine	Vrouwelijk, zacht
Cosette	Vrouwelijk, jong
Eponine	Vrouwelijk, expressief
Azelma	Vrouwelijk, helder

Prestaties

• circa 6x realtime op MacBook Air M4
• circa 200ms eerste chunk
• 2 CPU-kernen

---

TTS 1.6B (GPU)

Automatische keuze CUDA > MPS > CPU.

Vereisten

Onderdeel	Minimum	Aanbevolen
Python	3.10+	3.12
GPU VRAM	6GB	8GB+
RAM	8GB	16GB+
Schijf	4GB	8GB

Platformen

Platform	Backend	Notities
NVIDIA GPU	CUDA	Beste, bfloat16
Apple Silicon	MPS	float16
CPU	PyTorch	Langzaam, float32

Snel starten

cd examples/kyutai-tts-1.6b-server

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install PyTorch with CUDA (for NVIDIA GPUs)
pip install torch --index-url https://download.pytorch.org/whl/cu121

# Install dependencies
pip install -r requirements.txt

# Start server (auto-detects GPU)
python server.py

Server: http://localhost:8201.

Apparaat kiezen

# Auto-detect (CUDA > MPS > CPU)
python server.py

# Force specific device
python server.py --device cuda
python server.py --device mps
python server.py --device cpu

Test

curl http://localhost:8201/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model": "kyutai-tts-1.6b", "input": "Hello from the GPU!", "voice": "alba"}' \
  --output speech.wav

Stemmen

Alba MacKenna (CC BY 4.0):

Stem	Stijl
alba / alba-casual	Informeel
alba-merchant	Handelaar
alba-announcer	Omroeper

Expresso (CC BY-NC 4.0 - niet-commercieel):

Stem	Emotie
expresso-happy	Blij
expresso-sad	Verdrietig
expresso-angry	Boos

VCTK (CC BY 4.0):

• vctk-p225, vctk-p226, vctk-p227, vctk-p228

---

Stemklonen

Pocket TTS

# From local file
curl http://localhost:8200/v1/audio/voice-clone \
  -F "input=Hello from a cloned voice" \
  -F "reference_audio=@my_voice.wav" \
  --output cloned.wav

# From HuggingFace URL
curl http://localhost:8200/v1/audio/voice-clone-url \
  -H "Content-Type: application/json" \
  -d '{
    "input": "Hello world!",
    "voice_url": "hf://kyutai/tts-voices/alba-mackenna/casual.wav"
  }' \
  --output speech.wav

TTS 1.6B

Gebruik een Hugging Face-pad als voice:

curl http://localhost:8201/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kyutai-tts-1.6b",
    "input": "Custom voice synthesis",
    "voice": "hf://kyutai/tts-voices/vctk/p230.wav"
  }' \
  --output speech.wav

---

API-referentie

Spraakgeneratie

Endpoint: POST /v1/audio/speech

{
  "model": "kyutai-tts",
  "input": "Text to convert to speech",
  "voice": "alba",
  "response_format": "wav",
  "stream": false
}

Parameter	Type	Standaard	Beschrijving
model	string	varies	kyutai-tts of kyutai-tts-1.6b
input	string	required	Tekst, max 10.000
voice	string	alba	Naam of Hugging Face-pad
response_format	string	wav	Alleen wav
stream	boolean	false	Alleen Pocket
cfg_coef	float	2.0	Alleen 1.6B

Antwoord: audio/wav

OpenAI-stemaliassen

OpenAI	Pocket	1.6B
alloy	alba	alba
echo	marius	vctk-p225
fable	cosette	expresso-happy
onyx	javert	vctk-p226
nova	fantine	alba-announcer
shimmer	eponine	alba-merchant

Stemmenlijst

GET /v1/voices

Gezondheid

GET /health

---

Pluginconfiguratie

Pocket TTS

Instellingen > Plugins > Kyutai TTS, bestand plugins/kyutai-tts.json

{
  "id": "kyutai-tts",
  "name": "Kyutai TTS",
  "type": "tts",
  "endpoint": "http://localhost:8200/v1/audio/speech",
  "capabilities": {
    "tts": {
      "config": {
        "voices": [
          "Alba",
          "Marius",
          "Javert",
          "Jean",
          "Fantine",
          "Cosette",
          "Eponine",
          "Azelma"
        ],
        "default_voice": "Alba",
        "supports_streaming": true,
        "no_auth_required": true
      }
    }
  }
}

TTS 1.6B

Instellingen > Plugins > Kyutai TTS 1.6B, bestand plugins/kyutai-tts-1.6b.json

{
  "id": "kyutai-tts-1.6b",
  "name": "Kyutai TTS 1.6B",
  "type": "tts",
  "endpoint": "http://localhost:8201/v1/audio/speech",
  "capabilities": {
    "tts": {
      "config": {
        "voices": [
          "Alba",
          "Alba-Casual",
          "Alba-Merchant",
          "Alba-Announcer",
          "Expresso-Happy",
          "Expresso-Sad",
          "Expresso-Angry",
          "VCTK-P225",
          "VCTK-P226"
        ],
        "default_voice": "Alba",
        "supports_streaming": true,
        "no_auth_required": true
      }
    }
  }
}

---

Netwerktoegang

# Start server on all interfaces
python server.py --host 0.0.0.0

# Access from another machine
curl http://192.168.1.100:8200/v1/audio/speech ...

{
  "endpoint": "http://192.168.1.100:8200/v1/audio/speech"
}

---

Problemen oplossen

Modeldownload mislukt

# Set token for gated models
export HF_TOKEN=hf_...

CUDA-geheugen

Sluit andere GPU-apps, probeer cfg_coef=1.5 of gebruik Pocket.

Audiokwaliteit

Andere stem bij robotgeluid; lange tekst wordt automatisch gesplitst; model is geoptimaliseerd voor Engels/Frans.

MPS

RuntimeError: MPS backend error

python server.py --device cpu

---

Vergelijking met Qwen3-TTS

Functie	Pocket	1.6B	Qwen3-TTS
Parameters	100M	1.6B	0.6B-1.7B
GPU	Nee	Optioneel	Ja
Talen	Engels	EN/FR	10 talen
Klonen	Ja	Ja	Ja
Stemontwerp	Nee	Nee	Ja
Poort	8200	8201	8100

Kyutai voor Engels/eenvoud; Qwen voor meertalig en stemontwerp.

---

Bronnen

• Kyutai TTS - officieel
• Pocket TTS GitHub - CPU
• Delayed Streams Modeling - 1.6B
• Stemcollectie
• Modelkaart